An Introduction to Cassandra on Linux
- 1. Cassandra and Linux An Introduction Nick Bailey @nickmbailey nick@datastax.com Saturday, June 1, 13
- 3. ©2012 DataStax Analytics + Real Time 3 Big Data Saturday, June 1, 13
- 5. ©2012 DataStax Who is using it? 5 Saturday, June 1, 13
- 6. ©2012 DataStax 6 Saturday, June 1, 13
- 7. ©2012 DataStax Why do people like Cassandra? 7 Saturday, June 1, 13
- 9. ©2012 DataStax 9 http://techblog.netflix.com/2012/07/lessons-netflix-learned-from-aws-storm.html Saturday, June 1, 13
- 11. ©2012 DataStax 11 http://techblog.netflix.com/2011/11/benchmarking-cassandra-scalability-on.html Saturday, June 1, 13
- 13. ©2012 DataStax 13 http://vldb.org/pvldb/vol5/p1724_tilmannrabl_vldb2012.pdf Saturday, June 1, 13
- 14. ©2012 DataStax Multi Datacenter Support 14 Saturday, June 1, 13
- 15. ©2012 DataStax Hadoop Support 15 Saturday, June 1, 13
- 16. ©2012 DataStax Hadoop Support • Data Locality • Workload Partitioning 16 Saturday, June 1, 13
- 17. ©2012 DataStax Architecture - Cluster 17 Saturday, June 1, 13
- 18. ©2012 DataStax 18 Saturday, June 1, 13
- 19. ©2012 DataStax 19 Saturday, June 1, 13
- 20. ©2012 DataStax Architecture - Node 20 Saturday, June 1, 13
- 21. ©2012 DataStax Writes 21 Saturday, June 1, 13
- 22. ©2012 DataStax Writes 22 Saturday, June 1, 13
- 23. ©2012 DataStax Reads 23 Saturday, June 1, 13
- 24. ©2012 DataStax Reads 24 Saturday, June 1, 13
- 26. ©2012 DataStax Compaction • Periodically merge sstables • Multiple strategies • SizeTieredCompaction • LeveledCompaction 26 Saturday, June 1, 13
- 28. ©2012 DataStax Remember: Cassandra scales horizontally 28 Saturday, June 1, 13
- 29. ©2012 DataStax Memory 29 Saturday, June 1, 13
- 30. ©2012 DataStax Memory • More is better • Sweet spot: 16-64GB • Don’t give it all to the JVM • Generally no more than 8GB • Rest for page cache • Can run with less for quick testing 30 Saturday, June 1, 13
- 31. ©2012 DataStax CPU 31 Saturday, June 1, 13
- 32. ©2012 DataStax CPU • Cassandra is almost always IO bound • Sweet spot: 8 cores • Additional CPU required for: • compression • leveled compaction 32 Saturday, June 1, 13
- 33. ©2012 DataStax Disks 33 Saturday, June 1, 13
- 34. ©2012 DataStax Disks • SSDs are awesome, not required • Without SSDs: • At least 2 disks (commitlog, data) (more on that later) • Faster is better • Before Cassandra 1.2: ~500GB per node 34 Saturday, June 1, 13
- 35. ©2012 DataStax A Note on SSDs • Write Amplification • http://en.wikipedia.org/wiki/Write_amplification • Consumer grade SSDs are fine • See talk by Rick Branson for more • http://www.youtube.com/watch?v=zQdDi9pdf3I • http://www.slideshare.net/rbranson/cassandra-and- solid-state-drives 35 Saturday, June 1, 13
- 36. ©2012 DataStax Homogenous Nodes • Usually, keep nodes the same • Vnodes • Make heterogenous clusters easier • Added in version 1.2 36 Saturday, June 1, 13
- 38. ©2012 DataStax Disks 38 Saturday, June 1, 13
- 39. ©2012 DataStax 39 Saturday, June 1, 13
- 40. ©2012 DataStax Commit Log • Keep separate from data drives • Caveats • SSDs • Virtualized Environments 40 Saturday, June 1, 13
- 41. ©2012 DataStax Data Drives • Before Cassandra 1.2 • RAID0/RAID10 • Cassandra 1.2 • JBOD • Configuration options: stop/best_effort • XFS 41 Saturday, June 1, 13
- 42. ©2012 DataStax Note on SAN/NAS • Don’t use them • Cassandra is already distributed • SPOF • Cassandra is already IO bound 42 Saturday, June 1, 13
- 44. ©2012 DataStax Firewall • Ports: • 7000 - cluster communication • 9160 - client communication • JMX: • Unfortunately, the JMX protocol sucks • Ports 7199 and 1024+ for remote access • Solution: only access JMX locally 44 Saturday, June 1, 13
- 45. ©2012 DataStax Virtualized Environments (EC2) 45 Saturday, June 1, 13
- 46. ©2012 DataStax EC2 • Large/XLarge instances • Don’t use EBS • phi_convict_threshold • Don’t fix nodes, Replace them • DataStax provides an AMI 46 Saturday, June 1, 13
- 48. ©2012 DataStax Swap • Disable it • sudo swapoff --all • JVM swaps to disk, Cassandra explodes 48 Saturday, June 1, 13
- 49. ©2012 DataStax Limits • /etc/security/limits.conf • nofile • memlock • as 49 Saturday, June 1, 13
- 50. ©2012 DataStax NTP • Install it on • Cassandra Servers • Clients 50 Saturday, June 1, 13
- 51. ©2012 DataStax Monitor your cluster! • Cassandra exposes tons of metrics • Via JMX • Recently, more options available • DataStax OpsCenter • http://www.datastax.com/what-we-offer/products- services/datastax-opscenter • Or integrate with your own system 51 Saturday, June 1, 13
- 52. ©2012 DataStax Don’t use Windows • I’m not presenting at Texas Windows Fest • Technically supported • Not widely deployed • Reduced performance 52 Saturday, June 1, 13
- 53. ©2012 DataStax Resources • http://www.datastax.com/docs • #cassandra on freenode • http://www.planetcassandra.org • Mailing Lists • http://cassandra.apache.org to subscribe 53 Or... Saturday, June 1, 13
- 54. Come to the Summit! Ask me for a discount code (nick@datastax.com) June 11-12, 2013 San Francisco, CA http://www.datastax.com/company/news-and-events/events/ cassandrasummit2013 Saturday, June 1, 13
- 56. Questions? Saturday, June 1, 13